harnvo

Now, n_workers is 0 by default if the platform is win32. One is able to receive out of memory error if the memory of the disk of output dir does...

Hi, it's me again. I have found some redundancy in your code. These codes would not cause any error or bugs, but are not necessary and may reduce the run-time...

``` loss_D_real = 0 if D_main: if self.d_loss == "logistic": loss_D_real = torch.nn.functional.softplus(-real_logits) # -log(sigmoid(real_logits)) elif self.d_loss == "hinge": loss_D_real = torch.clamp(1.0 - real_logits, min = 0) elif self.d_loss ==...

Inconsistent evaluation of self.av

1

In stylegan2_pytorch.py, Trainer.generate_truncated(), self.GAN.S was used to evaluate the self.av in Trainer.truncate_style(). ``` @torch.no_grad() def truncate_style(self, tensor, trunc_psi = 0.75): S = self.GAN.S batch_size = self.batch_size latent_dim = self.GAN.G.latent_dim if...